Given how bad some of the models do on somewhat similar problems, I'm sure pelican is included in training set now.
Similar problems - given airplane outline and implementation constraints do painting scheme (constraints something like "it will be implemented using covering film, hence no gradients, no impossible cuts, not more than 2 colors on engine cowl, etc). Google Gemini is meh, but GPT models are just terrible, don't have Anthropic subscription at home, hence have not tested.
I did a little test that I like to do with new models: "I have rectangular space of dimensions 30x30x90mm. Would 36x14x60mm battery fit in it, show in drawing proof". GPT5 failed spectacularly.
Having written quite a bit of open API specs, I don't agree with you. Json is hard to read, yaml has own quirks, especially when you try to spilt it into parts.
Amazon also tries to invent own language for describing apis, so I guess they are not happy with open API too.
Anyway, without ability to generate code from spec, there is not much use from it. Code gen/nswagger/open API generator and others produce terrible code, at least for java/c#/typescript(there's 4.1k open issues for open API generator), using custom generators for codegen make problem less painful, but that is additional burden, I'm looking for better alternative, would be very interesting to see what they will do with code generation.
Tried couple queries that I've used lately for the job and results were meh, seems that getPayload was tokenized into get and Payload and that resulted into much not related stuff from sites that have nothing to do with programming. In code search in my opinion there needs to be subtle distinction when to do exact match, when not, even to keep syntax symbols, so that I could search for call usage, call usage with specific generic parameter, not definition, etc.